The regional comprehensive economic partnership through the eyes of AI

This post is sponsored by Isentia.

The media coverage of the regional comprehensive economic partnership (RCEP), a monumental deal among 15 Asia Pacific countries, has been interesting.

As media analysts, we looked into more than 800 articles and videos from the prominent news outlets across the 15 countries that signed the deal within the time period of 8 November to 20 November 2020.

Natural language processing uncovering the power dynamics in the region.

We employed state of the art AI models specialising in natural language processing (NLP), and contextualised them with our strategic viewpoint.

Sharing more about the AI analysis, Joseff Ymanuel Tan, Jr, data scientist at Isentia, noted: “NLP models work well for key information retrieval, specifically identifying proper nouns (entities). The entities are classified into several categories by the AI/NLP model.”

The first category we were interested in was the most frequently mentioned location names (country names, city names, etc). The image below provides a glimpse of the NLP model identifying the entities from an excerpt of an article.

The word cloud here shows the countries identified by the AI/NLP model. The bigger the country name is, the more the number of articles covered it. The US (appeared in 843 articles) and China (749 articles) were the top countries mentioned.

Sharing insights from the NLP analysis, Arisa Otsuka, analyst at Isentia, said: “This deal was seen as a significant victory for China as it reduces the impact of the ensuing trade war between the US and China by reducing China’s reliance on US trade [1]. The articles also mentioned the waning influence of the US as many allies rethink their economic ties with the US [2].”

One of those US allies being Japan (584) which was also the third most mentioned country. Some of the reasons why Japan was mentioned was its plan to sign military ties with Australia [3]; also Japan was covered as the one ensuring the protection of its local farmers by retaining tariffs on essential products such as rice, wheat, dairy, etc.[4]

Arun Elangovan, regional analytics manager of Isentia Asia, said: “At first glance of the country names, the AI is merely confirming the hypothesis that everyone would have regarding this deal, the 15 countries who signed the agreement are covered equally. But when we look at the results of the AI through a different lens, the message changes.”

While the previous word cloud shows how many articles the countries appeared in, the following word cloud demonstrates how many times the countries were mentioned across articles. This word cloud not only captures the essence of how prominently a country is covered, but also uncovers subliminal messages or in other words, the unsaid.

The articles more prominently mentioned China (4324 times) than any other country. The US (3171 times), Japan (1483) and Australia (1246) were some of the other countries also prominently mentioned. This could seed the idea that some countries are more influential than others, portraying the power dynamics in the region.

The media also extensively covered some countries that were not part of the deal. The first one being India (296 articles), as India was initially part of the RCEP and withdrew towards the end of last year. The coverage on India was mostly about how countries (especially Australia, Singapore and Vietnam) [5][6] were hopeful and keeping the door open for India to join the partnership in the future.

Japan considers India’s presence necessary to counter China’s economic weight [7]. The other country was Taiwan (66 articles). Taiwan was reportedly underplaying the significance of RCEP by pointing out that 70% of its exports to RCEP were already tariff free; instead, it is hopeful of joining the Trans-Pacific Partnership (CPTPP).[8]

Video AI hints towards a hopeful and prosperous future

The subliminal messaging in video coverage is less subtle as compared with articles. This is due to a video having two components: the visuals and audio/speech.

“Most of the news reporting on mainstream media state the facts through the audio/speech and show accompanying visuals. When we analysed the audio/speech, the narratives were very similar to the one we uncovered through the articles,” said Ma. Angelica Tatad, data scientist at Isentia.

But while analysing the visuals we found different narratives.

Francis De Leon, Jr, data scientist at Isentia, said the analysis of the video was “done through a state of the art neural network (AI) which can identify the objects shown in a video. Our analysts then group these object labels into relevant categories.”

The above image shows the AI (Google Video AI) identifying various objects shown in a video

There were some interesting groups of objects detected such as buildings and cityscapes. (The below graphs shows the various labels and the number of times they appeared in the videos).

The other group was transport and vehicles.

These groups of labels seem to indicate that the media is visually portraying this partnership as future-looking, and an opportunity for progress.

The other group of labels that was informative was the food group.

The media also showed various parts of the food supply chain, touching on one of the top priorities for countries especially in the Southeast Asia region – food security. This deal brings could potentially ease the tariffs on food import and export among the countries that are part of the deal.

Analysing the coverage around the RCEP through the eyes of the AI, revealed the media while largely emphasising the impact on the ongoing power dynamics, be it the dynamics between the US and its allies and China, also portrayed this deal as bringing a bright and prosperous future to the countries in the region. It also revealed there are some subtle/subliminal messages that lie underneath the obvious.